Enhancing Cardiac Arrest Prediction in Critically Ill Patients: A Sequence-Based Embedding Approach with Mamba

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Biomedical Informatics

Background:

• Cardiac arrest is a critical event in intensive care units (ICUs).
• Accurate and early prediction of cardiac arrest is crucial for timely intervention and improved patient outcomes.
• Existing prediction models may have limitations in performance and generalizability.

Purpose of the Study:

• To develop and validate a novel, robust model for predicting cardiac arrest in ICU patients.
• To evaluate the performance of a sequence-based embedding approach with a Mamba encoder for cardiac arrest prediction.
• To compare the proposed model against established prediction methods like NEWS and LightGBM.

Main Methods:

• Development of a novel sequence-based embedding model utilizing a Mamba encoder.
• Training and internal validation using ICU data from Seoul National University Hospital (SNUH).
• External validation using ICU data from Pusan National University Yangsan Hospital (PNUYH).

Main Results:

• The Mamba-based model achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.957 for internal validation.
• The model demonstrated strong external validation performance with an AUROC of 0.889.
• The proposed model outperformed both NEWS and LightGBM in cardiac arrest prediction accuracy.

Conclusions:

• The novel Mamba-based model shows significant potential for accurate and generalizable cardiac arrest prediction in ICUs.
• This approach offers a robust tool for early detection, enabling timely interventions and potentially improving patient survival rates.
• The findings support the integration of advanced AI models into clinical workflows for enhanced critical care management.